1. SIO 210: Data analysis methods L. Talley, Fall 2014 Reading: DPO Chapter 6 –Look only at the less mathematical parts –(skip sections on pdfs, least squares, EOFs and OMP except to note that the material is there) –i.e. you should know that these approaches exist but I don’t require you to know the mathematical expressions. –6.1, 6.2, 6.3.1, (6.3.3 for Section 2 students), 6.4, –6.5, (6.6.1 for Section 2), 6.6.2, 6.7.1, 6.7.2 1. Sampling and error 2. Basic statistical concepts 3. Time series analysis 4. Mapping 5. Filtering 6. Space-time data 7. Water mass analysis Talley SIO 210 (2014)

2. 1. Sampling and error: definitions Sampling (DPO Section 6.1) (review) –Synoptic sampling –Mean –Anomaly (difference between synoptic measurement and a mean Talley SIO 210 (2014)

3. 1. Sampling and error Error: (DPO Section 6.2) Measurement uncertainties (review) Precision: error within data set, or associated with instrument noise/resolution Accuracy: error relative to absolute standard Talley SIO 210 (2014)

4. 1. Sampling and error DPO Section 6.2 Random error: fluctuations (in either direction) of measured values due to precision limitations of the measurement device. Random error is quantified by the variance or standard deviation. (PRECISION) Systematic error (bias): offset, high or low, which cannot be determined through statistical methods used on the measurements themselves. An oceanographic example: Two or more technical groups measure the same parameters (e.g. temperature, nutrients or oxygen, etc). The mean values the groups obtain differ because of differences in methods, chemical standards, etc. Error can only be evaluated by comparison of the two sets of measurements with each other or with an absolute standard. (ACCURACY) Talley SIO 210 (2014)

5. Mean: Anomaly : x = x mean + x anomaly ; therefore x anomaly = x-x mean Variance: Standard deviation of the measurements:  Standard error: /√N 2. Basic statistical concepts: Mean, variance, standard deviation, standard error (DPO Section 6.3.1) Talley SIO 210 (2014)

6. 1. Sampling: mean and anomaly Climatology (mean): mean field based on large data base, usually covering many years Climatologyo of sea (Reynolds October Mean) Observed Field (Oct. 2012) Anomaly: Oct. 2012 Minus Reynolds October climatology http://www.pmel.noaa.gov/tao/ Talley SIO 210 (2014)

7. 2. Basic statistical concepts: Standard deviation vs. standard error Standard deviation is a measure of variability in the field that is measured. Standard error is a measure of how well the field is sampled. Standard deviation Standard error Salinity at 500 m Sampling Talley SIO 210 (2014) DPO Fig. 6.13

8. 3. Time series analysis (DPO 6.3.1, 6.3.3) A time series is a data set collected as a function of time Examples: current meter records, sea level records, temperature at the end of the SIO pier Some common analysis methods: A.Display the data (simply plot)! (always useful) B.Mean, variance, standard deviation, etc. (example not shown) C.Covariance and correlation of different time series with each other D.Spectral (Fourier) analysis Talley SIO 210 (2014)

9. DPO FIGURE 6.2 (a) Time series of temperature at Fanning Island (Pacific Ocean) from the NCAR Community Ocean Model. 3.A. Display time series Talley SIO 210 (2014)

10. 3.A. Display time series DPO FIGURE 6.6 Examples of time series plots: (a) property/time, (b) time series of profiles, (c) current speed and direction, and (d) stick diagram for data of (c).

11. 3.A. Display time series: Hovmöller diagram (time on one axis, space on the other) Equatorial Pacific sea level height anomaly from satellite (High sea level anomaly - El Nino!) (Low propagating west to east) Talley SIO 210 (2014)

12. 3.C. Time series analysis: covariance (DPO section 6.3.3) Covariance and correlation: integral of the product of two time series, can be with a time lag. (Autocorrelation is the time series with itself with a time lag.) Integral time scale: time scale of the autocorrelation (definitions vary) – crude definition might be at what time lag does the autocorrelation drop to zero (which is the decorrelation timescale) (First calculate the autocorrelation for a large number of time lags.) Degrees of freedom: total length of record divided by the integral time scale – that gives how many realizations of the phenomenon you have. Good to have about 10. Talley SIO 210 (2014)

13. Covariance : how two variables are related statistically. Correlation : covariance divided by the product of the (sample) standard deviations of the two variables. 3.C. Time series: covariance and correlation (DPO Section 6.3.3) Correlations have values from 0 to 1. How to tell if a correlation is significant? Talley SIO 210 (2014)

14. DPO FIGURE 6.2 (a) Time series of temperature at Fanning Island (Pacific Ocean) from the NCAR Community Ocean Model. (b) Autocorrelation normalized to a maximum value of 1 (biased estimate with averages divided by N). (c and d) Autocorrelation (unbiased estimate with averages divided by N– n). Source: From Gille (2005). Decorrelation time scale: 1 st 0 crossing. Integral time scale is a little longer (based on integral of the autocorrelation) 3.C. Time series: autocorrelation and timescales Talley SIO 210 (2014)

15. 3.C. Time series analysis: Correlation, degrees of freedom, confidence intervals (DPO Section 6.3.3) Confidence intervals: based on DoF and assumptions about statistical distribution 95% confidence interval: 95% probability that a value is within the standard error of the mean (times a t-test quantity that you look up). Talley SIO 210 (2014)

16. 3.C. Time series analysis: confidence intervals (DPO Section 6.3.3) DPO FIGURE 6.3 Example of time series with confidence intervals. Global ocean heat content (10 22 J) for the 0 to 700m layer, based on Levitus et al. (2005a; black curve), Ishii et al. (2006; full record gray curve and larger error bar), and Willis et al. (2004; darker gray after 1993 and shorter error bar). Shading and error bars denote the 90% confidence interval. Compare with Figure S15.15 seen on the textbookWeb site from Domingues et al. (2008) which uses improved observations

17. 3.C. Time series analysis: correlation example Example of correlation of different time series with each other North Atlantic Oscillation index (time series) Correlation with surface temperature and with surface pressure (done at each point, so each lat/lon gives a time series to correlate with NAO index) Talley SIO 210 (2014) Similar to DPO S15.2 and others S15 figures

18. 3.D. Time series analysis: spectral analysis (DPO Section 6.5.3) Spectral (Fourier) analysis of a time series is used to determine its frequency distributions. Underlying concept: any time series can be decomposed into a continuous set of sinusoidal functions of varying frequency. Spectrum: amplitude of each of the frequency components. Very useful for detecting things like tides, inertial motions, that have specific forcing frequencies. Talley SIO 210 (2014)

19. 3.D. Time series analysis: spectrum (DPO Section 6.5.3) DPO FIGURE 6.7 Example of time series, spectra, and spectral confidence intervals. (a) Velocity (cm/sec) stick plot, lowpassed at 100 hours, from 5 deep current meters at different depths on one mooring in the Deep Western Boundary Current in Samoan Passage (see Figure 10.16). The vertical direction is along the passage axis. (b) Spectra from the same current meters, offset by one decade. The 95% confidence intervals are shown at the bottom. Source: From Rudnick (1997).

20. 3.D. Time series analysis: spectrum (DPO Section 6.5.3) Spectra : Limitations are imposed by 1.sampling interval. The highest frequency that can be resolved in the spectrum(called the “Nyquist frequency”) is 1/2t where t is the sampling interval. Any energy at higher frequencies becomes “aliased” into lower frequencies, causing error in their amplitude estimates. 2.length of the total record. Generally should have about 10 realizations of a given frequency for the amplitude to be significant. Can prove difficult when trying to detect climate change for instance. This is related to “degrees of freedom”. Talley SIO 210 (2014)

21. 3.D. Time series analysis: Fundamental frequency and Nyquist frequency: sources of error in time series analysis (DPO Section 6.5.3) Sampling: N samples at time interval Δt. Length of time series is then T = NΔt Fundamental frequency –The lowest frequency that is resolved is » f = 1/T = 1/NΔt Nyquist frequency –The highest frequency that is resolved is » f = 1/2Δt Talley SIO 210 (2014)

22. 3.D. Time series: Nyquist frequency and aliasing (DPO Section 6.5.3) Aliasing: what happens to high frequencies that are above the Nyquist frequency cutoff? They are in the record, and are sampled, but not resolvable because their frequency content is not resolved They are aliased to a lower frequency. Talley SIO 210 (2014)

23. 4. Mapping - Least squares methods (DPO Section 6.3.4) How do we map data that are not regularly spaced (particularly in space)? “Least squares methods” are often used for mapping etc. A.Objective mapping B.Data assimilation (merging data and computer models) (no info given here) C.Inverse methods (finding “optimal” solution given sparse or incomplete data) (no info given here) Talley SIO 210 (2014)

24. 4.A. Mapping - Spatial sampling: Objective analysis (DPO Section 6.4) Many oceanographic data are collected in several spatial dimensions. Sample separations are very irregular. We want to interpolate these in an optimal way to a grid for plotting and comparison with other data sets. Objective analysis is a method for interpolating. It requires prior knowledge of the spatial correlation scales of the data to be mapped; since these would normally be calculated from the data themselves, they are often chosen in an ad hoc fashion. Talley SIO 210 (2014)

25. 4.A. Oceanographic sampling: objective mapping in the vertical (DPO Section 6.4.1) WOCE Pacific atlas (Talley, 2007) Discrete bottle samples: objectively mapped to uniform grid in vertical (10 m) and horizontal (10 km) Talley SIO 210 (2014)

26. 4.A. Oceanographic sampling: Vertical sections (DPO Section 6.4.1) Vertical coordinates: 1.Depth (pressure) 2.Potential density or neutral density (or theta) Talley SIO 210 (2014) DPO Figs. 4.11 and 9.17

27. 4.A. Oceanographic sampling: Mapping in the horizontal (DPO Section 6.4.2) DPO Fig. 6.4 Original station data: objectively mapped to a regular grid (NOAA/NODC) Different types of surfaces that are commonly used: 1.Constant depth 2.Isopycnal 3.“core layer” – here salinity maximum Talley SIO 210 (2014)

28. 5. Filtering (DPO Section 6.5.4) Time series or spatial sampling includes phenomena at frequencies that may not be of interest (examples: want inertial mode but sampling includes seasonal, surface waves, etc.) Use various filtering methods (running means with “windows” or more complicated methods of reconstructing time series) to isolate frequency of interest Example: Time series of a climate index with 1-year and 5-year running means as filters. Talley SIO 210 (2014) DPO Fig. 6.9

29. 6. Space-time sampling: empirical orthogonal functions etc (DPO Section 6.6.1) Underlying physical processes might not be best characterized by sines and cosines, especially in the spatial domain. They may be better characterized by functions that look like the basin or circulation geometries. Empirical orthogonal function analysis: let the data decide what the basic (orthogonal) functions are that add to give the observed values. Basic EOF analysis: obtain spatial EOFs with a time series of amplitude for each EOF. (The time series itself could be Fourier-analyzed if desired.) Talley SIO 210 (2014)

30. 6. Space-time sampling: empirical orthogonal functions (DPO Section 6.6.1) Davis, JPO, 1976 6 principal EOFs for sea level pressure anomalies in N. Pacific Talley SIO 210 (2014) DPO Fig. S6.3

31. 6. EOF example: Southern Annular Mode NAM Modes of decadal (climate) variability Talley SIO 210 (2014)

32. 7. Water mass analysis (DPO Section 6.7.2) Property-property relations (e.g., theta-S) Volumetric property-property Optimum multiparameter analysis (OMP) (Section 2 only) Talley SIO 210 (2014)

33. 7. Water mass analysis: potential temperature- salinity diagrams (DPO Section 6.7.2) Talley SIO 210 (2014) DPO FIGURE 6.14

34. 7. Water mass analysis: potential temperature- salinity diagrams (DPO Section 6.7.2) Talley SIO 210 (2014) DPO FIGURE 6.14 Potential temperature- salinity diagram used in several previous lectures looking at Atlantic properties

35. Potential temperature (ºC) versus salinity for (a) full water column, and (b) water colder than 10ºC. (c) Potential temperature versus oxygen for full water column. (d) Station location map. Colors indicate latitude range. Contours are potential density referenced to 0 dbar. Data are from the World Ocean Circulation Experiment (1988– 1997). This figure can also be found in the color insert. TALLEY Copyright © 2011 Elsevier Inc. All rights reserved DPO FIGURE 9.18 7. Water mass analysis: potential temperature- salinity diagrams (DPO Section 6.7.2)

36. Talley SIO 210 (2014) DPO FIGURE 6.15 7. Water mass analysis: More general property-property relations (DPO Section 6.7.2)

37. 7. Water mass analysis: volumetric potential temperature-salinity (DPO Section 6.7.2) Talley SIO 210 (2014) DPO 4.17b

38. 7. Optimum multiparameter analysis (OMP) (DPO Section 6.7.3) Use of water properties (T, S, oxygen, nutrients, other tracers): define source waters in terms of properties, then use a least squares analysis to assign every observation to a proportion of each source water Example of NADW vs. AABW: fractions of each (Johnson et al., 2008) DPO 14.15 Talley SIO 210 (2014)